clusboot {ClusBoot}

Performs bootstrap on a cluster analysis output

Description

Performs bootstrap on a cluster analysis output

Usage

clusboot(datmat, B = 1000, clustering.func = complete.linkage, ...)

Arguments

datmat

a data matrix or distance object which will be the input to the clustering function

B

number of bootstrap replicates

clustering.func

the function which will perform the clustering and output a vector of cluster memberships

...

more arguments to be passed to the clustering function

Details

Any R function performing cluster analysis can be specified in clustering.func although a wrapper function is typically needed to isolate only the vector output of cluster memberships. See ?complete.linkage as an example. Should users perfer to use alternative resamling schemes, other than the bootstrap, Hennig (2007) discuss a variety of options which could be accessed by specifying clustering.func = fpc.clusterboot. In addition, the sampling method is specified in the argument bootmethod and additional arguments for the function clusterboot in the package fpc must be given. Note that only the resampling facilities of clusterboot is utilised while the computation of proportions and silhouette widths remain unchanged. The output object of class clusboot will remain unchanged as only the resampling section of clusterboot is used.

Value

an object of class clusboot which is a list with the following components:

proportions

matrix of size nxn with cell ij containing the proportion of bootstrap replicates in which object i and object j clustered together.

clustering

a vector of length n containing the cluster membership of the n input objects.

sil

a vector of length the number of clusters containing the bootstrap-silhouette values for the clusters.

indv.sil

a vector of length n containing the bootstrap-silhouette values for the individual objects.

sil.order

a vector of length n containing the ordering of the n objects used by the functions boot.silhouette and boot.proportions to order objects in the same cluster adjacent and clusters in decreasing order of cluster tightness.

ave.sil.width

the overall stability of the clustering solution, obtained by averaging over the individual object bootstrap-silhouette values.

References

Hennig, C., 2007. Cluster-wise assessment of cluster stability. Computational Statistics & Data Analysis, 52(1), pp.258-271.

Examples

clusboot (scale(case.study.psychiatrist), B=100, k=6, clustering.func=complete.linkage)
library(fpc)
clusboot (scale(case.study.psychiatrist), B=100, k=6, clustering.func=fpc.clusterboot,
          clustermethod=hclustCBI, method="complete", bootmethod="subset", subtuning=10)